Research On MEC Computing Offloading And Resource Allocation Based On Deep Reinforcement Learning

Many promising applications in future mobile networks demand for more intensive computation,larger bandwidth and lower latency.However,the computing resources and battery capacity of current mobile devices are limited,which cannot meet the requirements of these applications.How to solve the contradiction between the resource-constrained mobile devices and the resource-hungry mobile applications has become a significant challenge for modern mobile network architecture.MEC is envisioned a new network architecture to address the above problems.MEC provides storage and computing resources at the edge of the mobile network.Task can be offloaded to the nearby cloud server for execution.The terminal device only needs to send the task request and receives result,so that the contradiction between the resource-constrained terminal device and the resource-hungry mobile applications can be effectively solved.For the wireless communication resources and the cloud computing resources,the dynamic offloading of tasks and the allocation of resources are a valuable issue.The research content of this paper is summarized as follows:First,a dynamic offloading and resource allocation scheme is proposed for a single cell multi-user scenario.Firstly,this paper establishes a model for network transmission and computation offloading,which aims to minimize the sum execution delay of all tasks.Secondly,a deep reinforcement learning scheme for task dynamic offloading and resource allocation is proposed.Finally,the simulation results show that the proposed algorithm can achieve better results under different simulation conditions.Second,considering the resource-constrained of the MEC service node,a multi-user,multi-cell access and resource allocation scheme is proposed,and each cell is connected to a high performance server.Firstly,this paper describes the network scenario of multi-user and multi-cell.Secondly,an optimization problem that aims to minimize the sum all delay of all tasks is established.Then,a deep reinforcement learning scheme based on task dynamic offloading,access and resource allocation is proposed.The final simulation results show that the proposed algorithm can reduce the delay while making full use of limited resources.